Camp Zero

Start while employed

How to ship an AI product after work

4 min read

Ship an AI product after work by picking one workflow a specific person already repeats, using a model to do a single step inside it, and putting a paid or booked pilot in front of five people within fourteen days. Evenings are enough for a narrow tool. They are a poor fit for a general assistant.

The AI products that fit a job are small on purpose. A general copilot needs a research team and a trust brand. A tool that turns one messy input into one deliverable your buyer already makes by hand can be scoped, prompted, and shown to a user on a weeknight.

Start from the workflow, then add the model

Write the job as a before-and-after. Before: a recruiter pastes a role brief into a doc and spends forty minutes drafting a scorecard. After: they paste the brief and receive a scorecard they can edit. The model does the draft. The human still decides. You did not build “AI for hiring.” You removed one painful step.

  • One user, described by role and context.
  • One input they already have.
  • One output they already know how to judge.
  • One place the output goes: a doc, an email, a spreadsheet, a ticket.

If you cannot name the input and the output, you are still browsing model demos. Keep a notes file of workflows you see at close range, using only information you are allowed to use. Your own freelance process is a fair source. Your employer’s private data is not.

A fourteen-day plan for five-hour weeks

  1. Days 1–2: write the offer sentence and the sample input/output. Show it to three people who do the job. Rewrite their objections into the spec.
  2. Days 3–6: make the thinnest path. A form, a prompt, a result on screen. Skip accounts and billing until someone asks to return.
  3. Days 7–8: sit with two users while they try it. Fix the confusing line, not the architecture.
  4. Days 9–12: add the smallest gate to a second use. A payment link, a calendar hold, or a manual invoice all count.
  5. Days 13–14: write the log. Who used it twice, what broke, whether next week is distribution or a fix.

That plan assumes the five-hour week. If a day job emergency eats a block, slide the plan. Do not add a second feature to feel productive.

Distribution while you are still employed

Your first users should be people you can email without borrowing your company’s list. Send the sample output, not a manifesto about AI. Offer to run their real input once, with their permission, and on your own machine. Ask if the draft saved them time they can feel. A yes with a number is the metric. A compliment is not.

Publish the build log where other AI builders will pressure you to show the artifact. Camp Zero’s house includes people shipping AI products after hours, alongside operators and creators. The point of the room is the next rep, which is also the point of building in public with a day job.

Add features only after five people have used the current one for a real task. Until then, your job is the workflow, the words on the page, and the next conversation.

Questions people ask

Do you need to train your own model?

Almost never for a first version. A clear workflow, a strong prompt, and a place to paste the result will teach you more than a training run. Custom training is a later project, after you know which errors users actually hate.

How do you handle customer data at night?

Ask before you put someone else’s material into a tool. Prefer sample data for the demo. Use accounts and settings you control, on your own device, and delete inputs you do not need to keep. If the workflow touches regulated data, get qualified advice before you sell it.

What is a good first price for a tiny AI tool?

Charge for the saved hour, not for the novelty. A manual invoice for a pilot is enough. Move to a subscription after people come back without you reminding them.